15-441 Computer Networking P2P Why P2P? • Scaling: system scales with number of clients, by definition • Eliminate centralization: • Eliminate single point of.

Download Report

Transcript 15-441 Computer Networking P2P Why P2P? • Scaling: system scales with number of clients, by definition • Eliminate centralization: • Eliminate single point of.

15-441 Computer Networking
P2P
Why P2P?
• Scaling: system scales with number of clients, by definition
• Eliminate centralization:
• Eliminate single point of failure
• Eliminate single point of control
• Use “spare” capacity
• Self-managing
15-441 S'10
Lecture 22: P2P
2
P2P example: file-sharing
• Lots of different P2P uses, but file sharing is common and
useful example
• Issues:
• Search: Find a file
• Fetch: Downloading the file
• Publish: inserting a new file
• Join: Maintaining network (e.g., dealing with faults, new
members, etc.)
15-441 S'10
Lecture 22: P2P
3
Approaches to P2P
•
•
•
•
Centralized
Flooding
Supernodes
Routing
• Structured
• Un-structured
15-441 S'10
Lecture 22: P2P
4
Centralized
•
•
•
•
•
(How is this P2P?)
Search: ask central server
Fetch: download from a peer
Publish: report to central server
Join: contact central server
15-441 S'10
Lecture 22: P2P
5
Flooding
• Search: ask everyone you know, then they ask, …
• How to limit search message traffic?
• Publish: implicit
• Join: let a peer know you exist
• Supernodes help with this, but in the end still has same
asymptotic properties
15-441 S'10
Lecture 22: P2P
6
Structured routing: e.g, Chord
• We saw how this might work with Chord
• Search: use hash to find node.
Makes none-exact searches hard
• Fetch: from single peer or ?
• Publish: implicit in joining
• Join: needs one peer in system
15-441 S'10
Lecture 22: P2P
7
Unstructured routing: E.g., Freenet
• Freenet is an example of an unstructured overlay using
key-based routing.
• Goals
• Censorship-resistance
• Anonymity: for producers and consumers of data
• Nodes don’t even know what they are storing
• Survivability: no central servers, etc.
• Scalability
15-441 S'10
Lecture 22: P2P
8
Big Idea: Keys as First-Class Objects
Keys name both the objects being looked up and the content itself
• Content Hash Key
• SHA-1 hash of the file/filename that is being stored
• Produces 160-bit digest
• Hash property: should be difficult to find a different file/filename that
hashes to same value
• File key
• Key is based on human-readable description of the file (SHA-1 hash)
• used to search for content
• Problem: same name might be selected for different files
• Encryption of files
• Documents can be encrypted: suggested key is unhashed humanreadable description of file
9
Publishing and Querying in Freenet
• Process for both operations is the same!
• Keys passed through a chain of proxy requests
• Nodes make local decisions about routing queries
• Queries have hops-to-live and a unique 64-bit ID
• Two cases
• Node has local copy of file
• File returned along reverse path
• Nodes along reverse path cache file
• Node does not have local copy
• Forward to neighbor “thought” to know of a key close to the
requested key
• First look up nearest key in routing table and forward request to
that neighbor, if that fails, try second nearest key, etc.
10
Satisfied Queries over time
15-441 S'10
Lecture 22: P2P
11
Routing Queries in Freenet
After success, node a creates a link in its routing
table for the key to node d.
Note: alternatively, any node on path from d to a,
12
Freenet Design
• Strengths
• Decentralized
• Anonymous
• Scalable
• Weaknesses
• Problem: how to find the names of keys in the first
place?
• No guarantee search will succeed
• No file lifetime guarantees
• No efficient keyword search
• No defense against DoS attacks
13
Freenet Security Mechanisms
• Encryption of messages
• Prevents eavesdropping
• Hops-to-live
• prevents determining originator of query
• prevents determining where the file is stored
• Hashing
• checks data integrity
• prevents intentional data corruption
14
P2P: summary
• Many different types: centralized, flooding, routing
• Issues:
• Failure mode: single point of failure? Search success?
• Flooding is onerous
• Network Topology different than overlay topology
• Nodes aren’t all the same
• Search can be hard
15-441 S'10
Lecture 22: P2P
15